Can I use CUDA 3.2 on a GTX 760 with the 335.23 drivers?

Accelerated Computing CUDA CUDA Setup and Installation
1

I have downloaded some code that insists that it can be built with (and only with) CUDA 3.2. Now that’s way out of date and I have a GTX 760 that likes CUDA v6. Can I install the 3.2 SDK and tools and run them with the current 335.23 drivers? I’ve been trying to do that and cannot get it to work. I tried downgrading to the 3.2-era drivers (v263.06) but the installation program says I do not have any valid hardware. (I guess it didn’t anticipate the 760!). I’ve got a windows 7 x64 box and I am trying to build the software with VS2010 and VC++ 10. It is an elaborate and complicated program that uses all manner of libraries (boost, f2c, lapack, Qt, etc). CUDA is used for both realtime graphics and the prior computation of some huge optimizations. The code not only insists on CUDA 3.2 but also claims that the CUDA parts can only by built with VC++ 9 compilers. So not only am I fighting with the CUDA versions but also trying to get VS2010 to use 2 compilers depending on the code. (Ah the joys of software written by PhD students!)

Am I just futzed here? What is the best way to solve this problem? I really do not want to have to spend much more time and effort trying to port the code from 3.2 to v 6. I’d just like to be able to use it.

So: is it even possible to use CUDA 3.2 with a GTX 760 with the 335.23 drivers? If I install the latest CUDA first (cuda_6.0.37_winvista_win7_win8.1_general_64.exe) and then follow that with the v 3.2 tool kit and SDK sample code, do I have any hope of making it work?

I’ve been working on this nonstop for the last week and have not been able to find any documentation related to such version compatibilities in CUDA.

Thanks

–AP

2

Try it both ways… try CUDA 5.5 or 6 with the drivers that you have, and if it doesn’t work, try it with 3.2. Just try it with the latest version of the respective libraries and see what happens… if something was deprecated, you should be able to figure out what the issue is and backpedal.

With regards to CUDA, code written in earlier SDKs should compile fine in the later versions… your trouble will most likely be with the external libraries and getting those configured correctly.